intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-23 11:03:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

[21/n] Remove Instruction Heap from enqueue path.

Browse Source 
- This removes Instruction Heap allocation from enqueue path
- Blocked path is handled as well
- Heap is no longer allocated on demand it is bind to kernelInfo.

Change-Id: I54545beceed3404ee0330a8bac2b0934944cac30
This commit is contained in:
Mrozek, Michal
2018-03-28 19:21:07 +02:00
parent f2b96fa508
commit 2be5934096
21 changed files with 49 additions and 277 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
unit_tests/command_queue/command_queue_hw_tests.cpp
View File

@@ -413,8 +413,7 @@ HWTEST_F(BlockedCommandQueueTest, givenCommandQueueWhichHasSomeUsedHeapsWhenBloc
mockKernelWithInternals.kernelHeader.KernelHeapSize = sizeof(mockKernelWithInternals.kernelIsa);
auto mockKernel = mockKernelWithInternals.mockKernel;
IndirectHeap::Type heaps[] = {IndirectHeap::INSTRUCTION, IndirectHeap::INDIRECT_OBJECT,
IndirectHeap::DYNAMIC_STATE, IndirectHeap::SURFACE_STATE};
IndirectHeap::Type heaps[] = {IndirectHeap::INDIRECT_OBJECT, IndirectHeap::DYNAMIC_STATE, IndirectHeap::SURFACE_STATE};
size_t prealocatedHeapSize = 2 * 64 * KB;
for (auto heapType : heaps) {
@@ -428,14 +427,8 @@ HWTEST_F(BlockedCommandQueueTest, givenCommandQueueWhichHasSomeUsedHeapsWhenBloc
DebugManager.flags.DisableResourceRecycling.set(false);
std::set<void *> reusableHeaps;
for (unsigned int i = 0; i < 5; ++i) {
for (unsigned int i = 0; i < 4; ++i) {
auto allocSize = prealocatedHeapSize;
//make sure that one of those allocations is larger so ISH can be recycled.
if (i == 4) {
allocSize = optimalInstructionHeapSize;
}
void *mem = alignedMalloc(allocSize, 64);
reusableHeaps.insert(mem);
memset(mem, 0, allocSize);
@@ -452,7 +445,6 @@ HWTEST_F(BlockedCommandQueueTest, givenCommandQueueWhichHasSomeUsedHeapsWhenBloc
userEvent.setStatus(CL_COMPLETE);
// make sure used heaps are from preallocated pool
EXPECT_NE(reusableHeaps.end(), reusableHeaps.find(pCmdQ->getIndirectHeap(IndirectHeap::INSTRUCTION, 0).getCpuBase()));
EXPECT_NE(reusableHeaps.end(), reusableHeaps.find(pCmdQ->getIndirectHeap(IndirectHeap::INDIRECT_OBJECT, 0).getCpuBase()));
EXPECT_NE(reusableHeaps.end(), reusableHeaps.find(pCmdQ->getIndirectHeap(IndirectHeap::DYNAMIC_STATE, 0).getCpuBase()));
EXPECT_NE(reusableHeaps.end(), reusableHeaps.find(pCmdQ->getIndirectHeap(IndirectHeap::SURFACE_STATE, 0).getCpuBase()));
@@ -479,8 +471,6 @@ HWTEST_F(BlockedCommandQueueTest, givenCommandQueueWhichHasSomeUsedHeapsWhenBloc
}
// expecting blocked command to be programmed indentically to a non-blocked counterpart
EXPECT_THAT(nonblockedCommandHeaps[static_cast<int>(IndirectHeap::INSTRUCTION)],
testing::ContainerEq(blockedCommandHeaps[static_cast<int>(IndirectHeap::INSTRUCTION)]));
EXPECT_THAT(nonblockedCommandHeaps[static_cast<int>(IndirectHeap::INDIRECT_OBJECT)],
testing::ContainerEq(blockedCommandHeaps[static_cast<int>(IndirectHeap::INDIRECT_OBJECT)]));
EXPECT_THAT(nonblockedCommandHeaps[static_cast<int>(IndirectHeap::DYNAMIC_STATE)],
@@ -506,12 +496,10 @@ HWTEST_F(BlockedCommandQueueTest, givenCommandQueueWhichHasSomeUnusedHeapsWhenBl
cl_event blockedEvent = &userEvent;
auto &ish = pCmdQ->getIndirectHeap(IndirectHeap::INSTRUCTION, 4096u);
auto &ioh = pCmdQ->getIndirectHeap(IndirectHeap::INDIRECT_OBJECT, 4096u);
auto &dsh = pCmdQ->getIndirectHeap(IndirectHeap::DYNAMIC_STATE, 4096u);
auto &ssh = pCmdQ->getIndirectHeap(IndirectHeap::SURFACE_STATE, 4096u);
auto ishBase = ish.getCpuBase();
auto iohBase = ioh.getCpuBase();
auto dshBase = dsh.getCpuBase();
auto sshBase = ssh.getCpuBase();
@@ -519,7 +507,6 @@ HWTEST_F(BlockedCommandQueueTest, givenCommandQueueWhichHasSomeUnusedHeapsWhenBl
pCmdQ->enqueueKernel(mockKernel, 1, &offset, &size, &size, 1, &blockedEvent, nullptr);
userEvent.setStatus(CL_COMPLETE);
EXPECT_EQ(ishBase, ish.getCpuBase());
EXPECT_EQ(iohBase, ioh.getCpuBase());
EXPECT_EQ(dshBase, dsh.getCpuBase());
EXPECT_EQ(sshBase, ssh.getCpuBase());